Floods are among the world’s most destructive natural hazards, making timely and reliable flood mapping essential for emergency response, damage assessment, and recovery planning. Because severe weather often limits the availability of cloud-free optical imagery, Synthetic Aperture Radar (SAR) has become a key technology for operational flood monitoring. Unlike optical sensors, SAR can acquire imagery day or night and through clouds, making Sentinel-1 SAR imagery an important data source for monitoring flood events.
Building an operational flood-monitoring workflow involves more than comparing two SAR images. Analysts must identify compatible pre- and post-flood acquisitions, account for acquisition geometry, manage multiple scenes covering the Area of Interest (AOI), and balance processing efficiency with cloud computing costs. These design decisions directly influence the reliability, reproducibility, and scalability of the final flood products.
This Python notebook, Automated Flood Monitoring with Sentinel-1 RTC Imagery, demonstrates how to implement an automated flood-monitoring workflow using ready-to-use Sentinel-1 Radiometrically Terrain Corrected (RTC) analysis-optimized imagery layer from ArcGIS Living Atlas, ArcGIS API for Python, and raster analytics. The workflow supports both fully automated processing and analyst-guided review, allowing the same methodology to accommodate different operational requirements. This article focuses on the reasoning behind the key design decisions for the notebook, discusses the trade-offs involved, and highlights the practices that make the workflow more reproducible, scalable, and reusable for operational flood monitoring.
Using flooding associated with Tropical Cyclone Idai near Beira, Mozambique, as an example, the notebook automatically identifies compatible Sentinel-1 image pairs, generates SAR change products, derives flood thresholds, refines flood masks and publishes the results as hosted imagery layers in ArcGIS Online. The workflow is designed to be reusable. By updating the flood event name, monitoring period, AOI, and processing mode, the same methodology can be applied to any flood event where and whenever Sentinel-1 imagery is available.
Table of Contents
- Step 1: Configure the Flood Event
- Step 2: Identify Compatible Image Pairs
- Step 3: Generate SAR Change Products
- Step 4: Detect Flooded Areas
- Step 5: Publish Flood Products
- Step 6: Review Results
- Extending the Workflow

Step 1: Configure the Flood Event
One design objective of this notebook is reusability. Rather than embedding event-specific values throughout the analysis, all configurable parameters are user-defined at the beginning of the workflow. This separation of configuration from processing logic makes it easy to apply the same methodology to different flood events without modifying the underlying code.
Connect to ArcGIS Online
Using analysis-ready imagery eliminates the need for local SAR preprocessing, allowing the notebook to focus on flood analysis rather than data preparation. Cloud-based processing also makes the workflow easier to scale from small study areas to larger operational analyses.
Define Flood Event Inputs and Processing Parameters
Every flood-monitoring workflow begins by defining the name, extent, and monitoring period for the flood event to be analyzed. These choices influence every subsequent stage of the workflow, including image selection, processing volume, and cloud computing costs. The notebook exposes a small set of high-level inputs so users can adapt the workflow to new flood events without modifying the underlying processing logic. The configuration also allows users to select the level of analyst involvement during processing.
Flood event extent and monitoring period parameters involve practical trade-offs. Increasing the monitoring period may capture more of the flood’s evolution and a larger AOI provides broader geographic coverage. However, increasing the AOI extent or monitoring period can increase the number of candidate images, which in turn increases processing time and ArcGIS Online credit consumption.
The example presented throughout this article uses flooding associated with Tropical Cyclone Idai near Beira, Mozambique, but these values simply configure the analysis. The underlying workflow remains unchanged.
Select a Processing Mode
One of the key design considerations in operational flood monitoring is determining how much analyst involvement is needed during processing. Fully automated workflows improve efficiency and consistency, particularly when analyzing multiple image pairs or monitoring recurring events. However, some situations benefit from additional analyst review, especially when evaluating image quality or managing cloud processing costs.
To accommodate these different requirements, the notebook supports two processing modes: AUTOMATIC and REVIEW. Both follow the same flood-detection methodology but differ in the level of analyst involvement.
AUTOMATIC mode prioritizes efficiency and repeatability by processing matched image pairs sequentially without requiring manual intervention. REVIEW mode allows analysts to inspect selected observations, evaluate intermediate results, and review estimated ArcGIS Online credit consumption before proceeding. At designated decision points, analysts can approve processing, skip individual image pairs, or stop the review, providing greater control over processing decisions and resource usage.
Create a Project Folder
A dedicated ArcGIS Online project folder is created to organize the outputs generated during the analysis. Keeping results grouped by flood event simplifies comparing multiple events and managing published imagery layers.
Load Sentinel-1 Imagery Layer and AOI
The Sentinel-1 RTC imagery provides analysis-ready SAR observations that require no additional preprocessing, while the AOI sets the processing bounds throughout the workflow, controlling credit usage and ensuring results share a consistent spatial extent.

Step 2: Identify Compatible Image Pairs
Reliable SAR change detection depends on comparing pre- and post-flood imagery acquired under similar viewing conditions. Differences in acquisition geometry can produce changes in radar backscatter that are unrelated to flooding, making image selection one of the most important factors affecting the quality of the final flood products.
Although selecting image pairs may seem straightforward, operational flood monitoring often involves dozens or hundreds of Sentinel-1 scenes acquired on different dates, relative orbits, orbit directions, and spatial footprints. Rather than relying on manual inspection, the workflow automatically applies a series of selection criteria to identify compatible image pairs. Automating this process improves reproducibility, reduces manual effort, and makes the workflow practical for repeated use across different flood events.
Query Sentinel-1 Scenes Intersecting the AOI
The workflow begins by retrieving Sentinel-1 scenes that intersect the AOI within the specified monitoring period. Building this pool of candidate observations automatically eliminates the need for manual image selection and provides a consistent starting point for selecting compatible image pairs.
The search window balances image availability with temporal relevance. A one-year lookback period typically provides sufficient pre-flood observations while reducing the likelihood of introducing long-term seasonal or land-cover changes. The user-defined monitoring period determines how long the workflow continues searching for post-flood imagery, making the same workflow suitable for both rapid flood response and longer-term monitoring.
Group Post-Flood Scenes into Acquisition Sets
Large AOIs are often covered by multiple scenes acquired during the same satellite pass. Rather than treating these scenes independently, the workflow groups observations that share the same acquisition date, relative orbit, and orbit direction into a single acquisition set. This creates a more consistent basis for comparing pre- and post-flood observations while simplifying the matching process.
Select Matching Pre-Flood Acquisition Sets
Selecting an appropriate pre-flood reference involves more than choosing the closest acquisition date. For SAR change detection, maintaining consistent acquisition geometry is essential, since differences in viewing geometry can produce backscatter changes unrelated to flooding.
For each post-flood acquisition set, the workflow searches the previous year for observations acquired from the same relative orbit and orbit direction. Candidate acquisitions are evaluated based on their AOI coverage, with preference given to the most recent pre-flood acquisition that satisfies the coverage requirements. If no candidate meets these requirements, the workflow selects the best available alternative, when one exists.
This approach prioritizes consistent viewing geometry while balancing spatial coverage and temporal proximity, improving the consistency of the resulting change products. By automatically selecting suitable image pairs, the workflow reduces manual intervention and supports reproducible processing across different flood events.
Prepare Matched Image Pairs for Processing
Once suitable image pairs are identified, they are retained throughout the remainder of the workflow to ensure that all subsequent raster processing uses the same observations. Locking the selected image pairs at this stage improves reproducibility and allows every downstream product to be traced back to a specific image pair.
Efficiently managing multiple image pairs is another important design consideration. Rather than retaining all imagery and intermediate raster-processing objects simultaneously, the workflow maintains lightweight metadata for the matched pairs and processes them sequentially. This approach reduces memory requirements and allows temporary resources to be released between pairs, improving processing efficiency and scalability.
Although image selection is automated, visual review provides an additional quality assurance opportunity before computationally intensive raster analytics begin. In REVIEW mode, analysts can inspect the selected observations and identify unsuitable image pairs before proceeding, helping avoid unnecessary processing and ArcGIS Online credit consumption.

Step 3: Generate SAR Change Products
With compatible image pairs identified, the workflow converts the selected SAR observations into quantitative measures of surface change. Separating change detection from flood classification is an intentional design decision that makes the workflow more flexible. The log-difference rasters provide a consistent representation of surface change, while the subsequent classification stage determines which changes are potentially associated with flooding.
Generate SAR Log-Difference Rasters
Flooding often reduces SAR backscatter because smooth water surfaces reflect much of the radar signal away from the sensor. To quantify these changes, the workflow generates a log-difference raster for each image pair. Values near zero indicate little or no measurable change, while negative values generally represent decreased backscatter associated with newly inundated areas. Positive values indicate increased backscatter that may result from vegetation changes, surface roughness, urban structures, or other land-surface changes.
Using a logarithmic change metric emphasizes relative changes while reducing the influence of SAR’s large dynamic range, providing a consistent representation of surface change across image pairs. These rasters are not flood maps themselves; instead, they serve as intermediate products that separate change detection from flood classification.
Clip Log-Difference Rasters to the AOI
Each log-difference raster is clipped to the AOI before further analysis. Processing only the required extent reduces the amount of data passed to subsequent raster analytics operations, improves efficiency, reduces ArcGIS Online credit consumption, and ensures that all subsequent outputs share a consistent spatial extent.
Review Log-Difference Rasters
Visual inspection of SAR change products is an important quality assurance practice, even when the underlying processing is automated. Widespread flooding typically appears as coherent regions of decreased backscatter, while stable areas remain close to zero. Reviewing these patterns helps determine whether the selected image pairs produced meaningful changes and provides an opportunity to identify unexpected artifacts before additional processing.
Both processing modes support visual inspection of the log-difference rasters. In AUTOMATIC mode, the results are displayed without interrupting processing, while REVIEW mode allows analysts to evaluate the results before proceeding with additional raster analytics operations.
Estimate Raster Analytics Credits for Log-Difference Outputs
Cloud-based raster analytics make large-scale processing practical, but they also consume ArcGIS Online credits. Estimating processing costs before creating hosted imagery layers helps users understand how factors such as AOI size, imagery extent, and the number of image pairs affect resource requirements.
The workflow estimates the credits required to save each log-difference raster before submitting the corresponding raster analytics operation. In AUTOMATIC mode, these estimates are reported without interrupting processing, prioritizing efficiency and repeatability. In REVIEW mode, analysts can evaluate the estimated costs before committing to credit-consuming operations.
This distinction reflects an important operational trade-off: automated processing minimizes interruptions, while analyst-guided review provides greater control over resource consumption.
Save Log-Difference Rasters
The workflow publishes clipped log-difference rasters as hosted imagery layers in ArcGIS Online. These intermediate products support the subsequent flood-classification stage by providing a consistent representation of surface changes before potential flooded areas are identified.
To manage storage efficiently, the log-difference layers are treated as temporary processing outputs and removed after successful publication of the corresponding flood masks. This approach reduces unnecessary storage consumption while preserving the final flood products for visualization, sharing, and further analysis.

Step 4: Detect Flooded Areas
The SAR change products quantify changes in radar backscatter, but not all changes are caused by flooding. Variations in vegetation, soil moisture, agricultural activity, urban surfaces, and inherent SAR speckle can all influence the radar backscatter, making flood detection more complex than simply identifying areas of change. To improve the reliability of flood detection, the workflow first identifies candidate flooded pixels using a histogram-derived threshold and then refines the result with neighborhood-based raster processing that suppresses isolated noise while preserving coherent flood patterns.
Determine Flood Thresholds from Log-Difference Histograms
Selecting an appropriate threshold is one of the most important decisions in SAR-based flood mapping because no single value performs well across all landscapes and flood events. Rather than using a predefined threshold, the workflow derives candidate thresholds directly from the histogram of each log-difference raster.
The notebook evaluates three commonly used methods:
- Triangle – Effective when flood pixels form a relatively small lower-value tail.
- Otsu – Maximizes the separation between two classes.
- Minimum – Identifies the valley between two peaks in bimodal histograms.
The workflow prioritizes the Triangle method and uses Otsu and Minimum as fallback methods when earlier methods cannot produce a valid threshold. This approach provides a consistent threshold-selection strategy while accounting for situations in which a particular method may not be applicable.
The selected threshold is a candidate flood-detection threshold rather than a validated measure of classification accuracy. Its suitability still depends on the characteristics of the imagery, landscape, and flood event.
Review Threshold Selection
Although threshold selection is automated, reviewing the histogram provides a useful quality assurance check. Displaying the candidate thresholds alongside the histogram helps users understand how the selected value relates to the distribution of observed backscatter changes.
This review also helps identify situations where unusual histogram shapes or local landscape characteristics may influence the automated threshold selection before additional processing is performed.
Both processing modes provide visibility into the threshold-selection results. In AUTOMATIC mode, the results are displayed without interrupting processing, while REVIEW mode allows analysts to examine the histogram and selected threshold before proceeding with flood classification.
Generate and Refine Flood Masks
After a threshold has been selected, the workflow converts the SAR change raster into a binary flood mask and refines it using neighborhood-based raster operations. This refinement reduces isolated detections caused by SAR speckle while preserving larger, spatially coherent flood patterns.
An intentional design choice is to represent flooded pixels with a value of 1 while preserving all non-flood areas as NoData. This allows the published flood masks to overlay cleanly on basemaps, Sentinel-1 imagery, and other operational datasets without obscuring the underlying information. This transparent representation also simplifies comparison with reference imagery, infrastructure, and other geospatial data during emergency response and post-event assessment.
Note: Like any automated SAR classification, the results should be interpreted alongside other available information. Permanent water bodies, smooth bare soil, radar shadow, and some agricultural surfaces can produce radar responses similar to flooding. Whenever possible, the flood masks should be validated using reference imagery, local knowledge, or ancillary datasets before being used for operational decision-making.

Step 5: Publish Flood Products
The workflow publishes the refined flood masks as hosted imagery layers in ArcGIS Online. Publishing transforms the analysis results into persistent data products that can be visualized, shared, and incorporated into other GIS applications without rerunning the workflow. Integrating publishing into the workflow also promotes consistency by automatically applying a standardized naming convention and preserving the relationship between the outputs and their corresponding image pairs.
Estimate Raster Analytics Credits for Publishing Flood Masks
Before publishing the final flood masks, the workflow estimates the ArcGIS Online credits required to create each hosted imagery layer. Reviewing these estimates helps users understand the resource requirements and balance analysis objectives with available credits.
Estimating costs separately for intermediate log-difference rasters and final flood masks provides greater transparency into the resources required at different processing stages. This allows analysts to assess the cost of generating the final products independently of the preceding change-detection operations.
In AUTOMATIC mode, processing continues after reporting the estimates, while REVIEW mode provides an opportunity to evaluate the expected costs before publishing the final flood masks.

Tip: When adapting the workflow to a new flood event or testing workflow modifications, start with a smaller AOI or a shorter monitoring period. This reduces processing time and credit consumption while validating the workflow.

Save Final Flood Mask Layers
After reviewing the estimated processing costs, the workflow publishes each refined flood mask as a hosted imagery layer in ArcGIS Online. Publishing preserves the final products independently of the notebook, making them available for visualization, sharing, and use in downstream GIS workflows without repeating the analysis.
Report Actual Raster Analytics Credit Usage
Once processing is complete, the workflow retrieves the actual ArcGIS Online credits consumed by each raster analytics task. Comparing estimated and actual credit usage helps users better understand the computational cost of the workflow and provides useful guidance when planning future analyses.

Step 6: Review Results
Generating a flood map is not the end of the analysis. Before the results are shared or used for decision-making, they should be reviewed to verify that the mapped flood extent is consistent with the underlying SAR observations. This final step focuses on quality assurance by improving visualization, providing quick access to the published imagery layers, and supporting visual interpretation of the results.
Apply Flood Symbology
The workflow applies a consistent symbology that displays flooded pixels in blue while leaving non-flood areas transparent (NoData). Using the same visualization for every output makes flood extents easier to interpret and compare across image pairs and flood events, while allowing the results to be viewed over basemaps, Sentinel-1 imagery, or other operational datasets without obscuring the underlying information.
Display Output Item Links
The workflow provides direct links to each published imagery layer in ArcGIS Online, allowing users to quickly access the outputs, review metadata, configure sharing settings, or incorporate the layers into web maps, dashboards, and other GIS applications.
Review Final Flood Masks
The final flood masks should be interpreted alongside the SAR change products and, whenever possible, other reference information. Comparing these datasets helps confirm that mapped inundation is consistent with the observed changes in radar backscatter while providing an opportunity to identify unexpected artifacts before the results are used operationally.
When multiple compatible image pairs are available, the workflow can also support temporal monitoring by producing a consistent series of flood masks that capture how inundation evolves through time. As new Sentinel-1 observations become available, the workflow can be rerun using the same methodology to monitor flood progression or recession.
Processing efficiency is another important consideration when working with multiple image pairs. The workflow processes pairs sequentially and releases temporary processing resources when they are no longer needed. This reduces memory requirements while maintaining consistency across acquisitions.
Both processing modes support visual inspection of the results. In AUTOMATIC mode, intermediate and final results are displayed without interrupting processing, while REVIEW mode allows analysts to examine the results before proceeding to subsequent image pairs.
Note: Like any automated SAR-based classification, the results have limitations. Permanent water bodies, smooth bare soil, radar shadow, and some agricultural surfaces can produce radar responses similar to flooding. Depending on the application, the flood masks may be further refined using ancillary datasets such as permanent water layers, land-cover classifications, or digital elevation models.
Review the Processing Summary
Reproducibility involves more than applying the same analytical methods. It also requires a clear understanding of which observations were processed and what decisions were made. To support this, the workflow consolidates processing information for all matched image pairs into a summary that includes acquisition dates, source image identifiers, processing status, selected thresholds, and estimated credits.
This summary provides a consistent way to review processing outcomes and trace individual flood products back to their source observations. It is particularly valuable when processing multiple image pairs, where some pairs may be completed, skipped, or left pending during analyst-guided review.
By bringing this information together, the workflow improves transparency and reproducibility without requiring users to inspect every intermediate processing step. It also provides a useful reference for reviewing results and managing outputs across different flood events.
This article introduced the design principles behind an automated flood-monitoring workflow built with Sentinel-1 imagery layer from Living Atlas, ArcGIS API for Python, and raster analytics. While the accompanying notebook demonstrates the implementation, this article focused on the design decisions that make the workflow dependable, reusable, and scalable for operational flood monitoring.
Although the workflow is demonstrated using the 2019 flooding associated with Tropical Cyclone Idai near Beira, Mozambique, it is designed as a reusable framework rather than an event-specific solution. By updating the flood event name, monitoring period, and AOI, the same methodology can be applied to any flood event where Sentinel-1 imagery is available to monitor flood progression and compare observations through time. Its modular design also makes it straightforward to incorporate additional datasets, alternative classification approaches, or application-specific refinements as operational requirements evolve.

Extending the Workflow
One of the strengths of this workflow is its modular design. Each stage—from image selection and change detection to thresholding and publishing—can be adapted without changing the overall framework. Potential extensions include:
- Permanent water masking – Incorporate datasets such as the Joint Research Centre (JRC) Global Surface Water dataset to reduce confusion between persistent water bodies and newly inundated areas.
- Slope filtering – Use a digital elevation model (DEM) to remove unlikely flood detections in steep terrain.
- Additional SAR information – Evaluate the combined use of VV and VH polarizations to determine how different polarization combinations influence flood mapping results.
- Alternative thresholding approaches – Compare additional histogram-based or locally adaptive thresholding methods.
- Alternative change-detection approaches – Extend the current log-difference methodology by incorporating direct thresholding of pre- or post-flood imagery, or by combining absolute SAR backscatter values with change-detection products to further refine flood classification.
- Application-specific customization – Adapt the image-selection logic, matching criteria, thresholding methods, flood-mask refinement, processing controls, visualization, and output products to meet specific operational or research requirements.
These examples illustrate how the workflow can evolve while preserving its core design principles: automated image selection, reproducible processing, flexible analyst oversight, and scalable cloud-based analysis.
Additional resources
You can find more tutorials in the tutorial gallery.
Explore additional raster analysis blog tutorials using Living Atlas imagery content.
Learn more about SAR in Guide: Fundamentals of Synthetic Aperture Radar.
Learn more about the ArcGIS API for Python.
Explore the Deep Learning Libraries Installers for ArcGIS.
Acknowledgements
Sentinel-1 RTC layer: Copernicus Sentinel-1 data from European Space Agency, sourced from the Microsoft Planetary Computer and accessed via Living Atlas.
Article Discussion: